Extended Data Fig. 4: Benchmarking of SLMHRD.
From: The transcriptomic architecture of common cancers reflects synthetic lethal interactions
a. Benchmarking of SLMHRD against three other RNA-based DNA damage response deficiency signatures in six independent studies (14 datasets). Heatmap showing significance of association between patients’ outcome data and signature score-derived patient risk groups (for prognostic models), or, difference in signature scores between clinical response following neoadjuvant chemotherapy (for predictive models; residual disease or pathological complete response). Significance is reported as trend P value for prognostic models and a P value from generalised linear models for predictive associations (models adjusted for hormone receptor status for I-SPY2, Hatzis and Horak datasets; two-sided Wald-test). Number in cells show the rank (across a row) of each signature compared to other signatures for each dataset. Cumulative ranking was calculated using the geometric mean of individual ranks and shown as a bar plot, with smallest geometric mean indicating best overall performance.
